Poster #80: Visnu Chowdhury MS

Rational domain segmentation improves discovery of phosphoinositide-binding bacterial effectors via 3D structural homology

Visnu P. Chowdhury (MS)¹, Abby E. Richardson (BS)², Abigail E. Bolt (PhD)², M. Ramona Neunuebel (PhD)², Karl R. Schmitz (PhD)¹,²,³ Affiliations: ¹Center for Bioinformatics and Computational Biology, University of Delaware, Newark, DE ²Department of Biological Sciences, University of Delaware, Newark, DE ³Department of Chemistry & Biochemistry, University of Delaware, Newark, DE

Legionella pneumophila, the causative agent of Legionnaires’ Disease, is a common model system for studying intracellular bacterial pathogens. Legionella survives within host cells by injecting over 350 “effector” proteins that remodel host cell membranes and disrupt the normal endocytic-lysosomal pathway. Some Legionella effector proteins localize to specific vesicular compartments by interacting with the phosphoinositide (PIP) lipids that mark organellar identity. Recent studies by Bolt et al. uncovered novel bacterial PIP-binding effectors by identifying structural homology between validated PIP-binders and AlphaFold structural predictions of proteins in the VFDB (Virulence Factor Database). However, the structural predictions in this analysis were generated from 200-amino-acid segments of VFDB proteins, risking domain fragmentation and missing valid PIP-binding domains. We hypothesize that rational domain segmentation of full-length proteins will recover additional novel PIP-binding effectors. Methods: We have constructed a computational pipeline to systematically extract and analyze protein domains from AlphaFold-predicted structures. Virulence factor database entries were correlated with UniProt identifiers, which were used to retrieve existing AlphaFold2 predictions. Predicted Aligned Error (PAE) matrices and graph-based clustering (Louvain and Markov Clustering algorithms) of residue contact maps were used to identify protein domains, which were further validated using DSSP to identify secondary structure boundaries, as well as the domain segmentation tools DPAM and Merizo, while prioritizing consensus boundaries across tools. Extracted domains were compared against validated PIP-binding effectors using flexible alignment algorithms to identify structurally homologous domains. Significance & Conclusion: This pipeline provides a robust, generalizable, domain-agnostic approach for identification of structural homologs of functionally characterized protein queries, which can potentially find functionally conserved proteins that share a common 3D motif despite having poor sequence conservation.